A non-iterative approach to initial region estimation applied to color image segmentation

A non-iterative segmentation approach is developed to generate a fast initial estimation of the layout of the different color textures presented in the original image mainly based on hypothesis testing. Most of the proposed methods have a tremendous computational burden which make them difficult to be implemented in a real-time working processor. We will compare our method with a known iterative clustering algorithm that guides to similar results with much higher computational cost. We present two examples that show similar results and compare the computational cost for each case. Spotty resemblance caused by pixel oriented decision is diminished in both cases by modeling regions as Markov Random Fields.